Model

Claude 3.5 Haiku

CLOSED

Anthropic · Claude 3.5 family · released Oct 22, 2024

Fast low-cost Claude 3.5 tier that matched Claude 3 Opus on many benchmarks; retired Feb 2026.

ReasoningCodingVisionFunction callingTool useAgentic

1041.8

Elo · rank #205

Parameters

Undisclosed

Active params

—

Context

200K tokens

Architecture

Dense transformer (parameter count undisclosed)

License

Proprietary

Languages

—

API price (in/out)

$0.8 / $4

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath5.3%#150

best: GPT-5.2 · 100.0%

Arena EloHuman preference1324#33

best: Claude Fable 5 · 1505

BIG-Bench HardReasoning86.6%#12

best: ERNIE 4.5 300B-A47B · 94.3%

DROPReasoning83.1#18

best: Hunyuan-T1 · 93.1

FrontierMath (Tiers 1-3)Math0.3%#14

best: GPT-5.6 · 89.0%

GDPval-AAAgents500#36

best: Claude Fable 5 · 1932

GPQA DiamondReasoning41.6%#212

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning3.0%#136

best: Claude Sonnet 5 · 57.4%

HumanEvalCoding88.1%#30

best: Claude Opus 4.5 · 99.4%

IFBenchReasoning43.0%#42

best: MiniMax M3 · 83.0%

IFEvalReasoning85.9%#52

best: Gemma 4 26B A4B · 98.5%

MATH-500Math69.2%#97

best: GPT-5 · 99.4%

MGSMMath85.6%#23

best: OpenAI o4-mini · 93.7%

MMLU-ProKnowledge65.0%#108

best: Claude Fable 5 · 91.5%

MMLUKnowledge80.3%#89

best: OpenAI o3 · 92.9%

MMMU-ProVision46.0%#47

best: Claude Opus 4.7 · 85.5%

SWE-bench VerifiedCoding40.6%#104

best: Claude Fable 5 · 95.0%

tau-benchAgents51.0%#21

best: Claude Opus 4.1 · 82.4%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$0.8

Output / M tok

Claude 3.5 family

Elo progression across releases

Claude 3.5 SonnetJun 20241294.7 Claude 3.5 HaikuOct 20241041.8 Claude 3.5 Sonnet (New)Oct 20241357.3

Resources

DOCSplatform.claude.com

API price $0.8/$4 · each benchmark row carries its own source badge (see methodology)